| name | notebooklm-research |
| description | Source-grounded research workflow for NotebookLM URLs, notebook-backed documentation queries, and MCP-driven follow-up analysis. Use when a task includes a NotebookLM URL, asks to query notebook contents, register a notebook, or persist grounded notes before implementation. |
NotebookLM Research
Trigger
- User provides a NotebookLM URL.
- User asks to "query NotebookLM", "use notebook docs", or "research this notebook before coding".
- The task needs source-grounded answers before implementation.
Read First
../../docs/ref/NOTEBOOKLM-MCP-RUNBOOK.md
../CLAUDE.md
../../docs/ref/notebooklm/6a8491af-e793-4574-8182-9ab2511fbbe3.md when the provided notebook matches the GameLab shared notebook
Input Context
- Notebook URL or notebook ID
- Research question or implementation target
- Expected output path if the user wants notes persisted
Do
- Confirm the
notebooklm MCP server is configured and the toolset includes get_health, list_notebooks, add_notebook, select_notebook, and ask_question.
- If authentication is missing, stop and ask the user to complete NotebookLM browser login before claiming any notebook content.
- Register the notebook if it is not already in the library.
- Select the notebook before asking implementation questions.
- Ask an initial question set:
- notebook overview
- concepts relevant to the current task
- constraints or failure modes
- implementation details still missing
- Ask follow-up questions until the notebook answer covers the task, or explicitly states the information is unavailable.
- Persist a concise summary to
../../docs/ref/notebooklm/<notebook-id>.md when the task needs a reusable reference.
- Append a machine-readable entry to
../../memory/notebooklm-runs.ndjson with notebook ID, question scope, and status.
Do Not
- Do not pretend NotebookLM content was retrieved if authentication or sharing blocks access.
- Do not replace repo source-of-truth docs with notebook notes.
- Do not store secrets, tokens, or copied raw proprietary content in the repo.
Validation
- NotebookLM MCP is configured.
- Notebook selection is explicit before question asking.
- Follow-up questions were used when the first answer was incomplete.
- Any saved note includes source URL, sync status, and open gaps.
Output
[notebooklm-research]
- notebook: <id or url>
- status: success | blocked-auth | blocked-sharing | partial
- questions: <count>
- saved_ref: <path or none>
- log_appended: yes | no
- open_gaps: <summary>